ViewVC Help
View File | Revision Log | Show Annotations | Revision Graph | Root Listing
root/i-scream/projects/cms/source/host/ihost-perl/ihost.pl
(Generate patch)

Comparing projects/cms/source/host/ihost-perl/ihost.pl (file contents):
Revision 1.6 by pjm2, Fri Jan 26 17:47:18 2001 UTC vs.
Revision 1.49 by tdb, Sat May 18 18:15:56 2002 UTC

# Line 1 | Line 1
1   #!/usr/bin/perl -w
2  
3 + #
4 + # i-scream central monitoring system
5 + # Copyright (C) 2000-2002 i-scream
6 + #
7 + # This program is free software; you can redistribute it and/or
8 + # modify it under the terms of the GNU General Public License
9 + # as published by the Free Software Foundation; either version 2
10 + # of the License, or (at your option) any later version.
11 + #
12 + # This program is distributed in the hope that it will be useful,
13 + # but WITHOUT ANY WARRANTY; without even the implied warranty of
14 + # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 + # GNU General Public License for more details.
16 + #
17 + # You should have received a copy of the GNU General Public License
18 + # along with this program; if not, write to the Free Software
19 + # Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
20 + #
21 +
22   # -----------------------------------------------------------
23   # Perl i-scream Host.
24 + # http://www.i-scream.org.uk
25   #
26   # An all-in-one script to act as an i-scream host on
27 < # a typical Unix/Linux box.  You may adapt the data-gathering
8 < # methods as you see fit.
9 < #  - pjm2@ukc.ac.uk
27 > # a typical Unix/Linux box.
28   #
29   # $Author$
30   # $Id$
# Line 31 | Line 49 | use vars qw (
49               $tcp_port
50               $filter_addr
51               $file_list
52 +             $fqdn
53 +             $pidfile
54 +             $retry_wait
55 +             $ostype
56 +             $key
57 +             @data
58              );
59  
60   if (@ARGV != 2) {
# Line 41 | Line 65 | $filter_manager_addr = $ARGV[0];
65   $filter_manager_port = $ARGV[1];
66  
67   $seq_no = 1;
68 + $retry_wait = 60;
69  
70 + # work out our platform, if we can.
71 + $ostype = `uname -s`;
72 + chomp $ostype;
73 + $ostype = "unknown" if not defined $ostype;
74 +
75 + # write our PID to a file
76 + # use home dir by default
77 + #$pidfile = $ENV{"HOME"};
78 + # or drop it in /var/tmp if we can't find HOME
79 + $pidfile = "/var/tmp" if not defined $pidfile;
80 + $pidfile .= "/.ihost.pid";
81 + &write_pid();
82 +
83   &tcp_configure();
84 + &send_tcp_heartbeat();
85   &send_udp_packet();
86  
87   $last_udp_time = time;
88   $last_tcp_time = time;
89   while (1) {
90 <    if (time >= $last_udp_time + $udp_update_time) {
90 >    my($time) = time;
91 >    if ($time >= $last_udp_time + $udp_update_time) {
92          &send_udp_packet();
93 <        $last_udp_time = time;
93 >        $last_udp_time = $time;
94      }
95 <    if (time >= $last_tcp_time + $tcp_update_time) {
95 >    if ($time >= $last_tcp_time + $tcp_update_time) {
96          &send_tcp_heartbeat();
97 <        $last_tcp_time = time;
97 >        $last_tcp_time = $time;
98      }
99 <    `sleep 1`;
99 >    my($next_udp) = $udp_update_time - $time + $last_udp_time;
100 >    my($next_tcp) = $tcp_update_time - $time + $last_tcp_time;
101 >    my($delay);
102 >    if ($next_udp < $next_tcp) {
103 >        $delay = $next_udp
104 >    }
105 >    else {
106 >        $delay = $next_tcp;
107 >    }
108 >    sleep $delay;
109   }
110  
111 + # we'll probably never get here...
112 + `rm -f $pidfile`;
113   exit(0);
114  
115 +
116 + #-----------------------------------------------------------------------
117 + # wait_then_retry
118 + # Waits for the period of time specified in $retry_wait, then attempts
119 + # to reconfigure with the server.
120 + #-----------------------------------------------------------------------
121 + sub wait_then_retry() {
122 +    print "Will retry configuration with filter manager in $retry_wait seconds.\n";
123 +    sleep $retry_wait;
124 + }
125 +
126 +
127 + #-----------------------------------------------------------------------
128 + # tcp_configure
129 + # Establishes a TCP connection to the specified i-scream filter manager.
130 + # The host then requests details from the server, such as the intervals
131 + # at which to send UDP packets.
132 + #-----------------------------------------------------------------------
133   sub tcp_configure() {
134      
135 <    my($sock) = new IO::Socket::INET(
136 <                                     PeerAddr => $filter_manager_addr,
137 <                                     PeerPort => $filter_manager_port,
138 <                                     Proto => 'tcp'
139 <                                    ) or die "Could not perform configuration via TCP: $!\n";
135 >    while (1) {
136 >        my($sock) = new IO::Socket::INET(
137 >                                         PeerAddr => $filter_manager_addr,
138 >                                         PeerPort => $filter_manager_port,
139 >                                         Proto => 'tcp'
140 >                                        ) or die "Cannot connect!";
141 >        if (!defined $sock) {
142 >            print "IHOST ERROR: Could not connect to $filter_manager_addr:$filter_manager_port.\n";
143 >            print "Please check that there is an i-scream server at this address.\n";
144 >            wait_then_retry();
145 >            next;
146 >        }
147  
148 <    die "Could not connect to the i-scream filter manager: $!\n" unless $sock;
149 <
74 <    # Now run through the configuration process.
75 <    my($response);
148 >        # Now run through the configuration process...
149 >        my($response);
150      
151 <    print $sock "STARTCONFIG\n";
152 <    $response = <$sock>;
153 <    if (!chop $response eq "OK") {
154 <        print "The i-scream server rejected the STARTCONFIG command.  Terminated.";
155 <        exit(1);
156 <    }
151 >        print $sock "STARTCONFIG\n";
152 >        $response = <$sock>;
153 >        if ($response && !($response eq "OK\n")) {
154 >            print "The i-scream server rejected the STARTCONFIG command.\n";
155 >            close($sock);
156 >            wait_then_retry();
157 >            next;
158 >        }
159  
160 <    print "Config started okay.\n";
160 >        print "Config started okay.\n";
161      
162 <    print $sock "LASTMODIFIED\n";
163 <    $response = <$sock>;
164 <    chop $response;
165 <    $last_modified = $response;
162 >        print $sock "LASTMODIFIED\n";
163 >        $response = <$sock>;
164 >        if (!$response || $response eq "ERROR\n") {
165 >            print "The i-scream server did not provide the LASTMODIFIED value.\n";
166 >            close($sock);
167 >            wait_then_retry();
168 >            next;
169 >        }
170 >        chomp $response;
171 >        $last_modified = $response;
172      
173 <    print "Config last modified: ". scalar localtime $last_modified . "\n";
173 >        print "Config last modified: ". (scalar localtime $last_modified/1000) . "\n";
174  
175 <    print $sock "FILELIST\n";
176 <    $response = <$sock>;
177 <    chop $response;
178 <    $file_list = $response;
175 >        print $sock "FILELIST\n";
176 >        $response = <$sock>;
177 >        if (!$response || $response eq "ERROR\n") {
178 >            print "The i-scream server did not provide a configuration file list.\n";
179 >            close($sock);
180 >            wait_then_retry();
181 >            next;
182 >        }
183 >        chomp $response;
184 >        $file_list = $response;
185  
186 <    print "File list obtained: $file_list\n";
186 >        print "File list obtained: $file_list\n";
187  
188 <    print $sock "UDPUpdateTime\n";
189 <    $response = <$sock>;
190 <    chop $response;
191 <    $udp_update_time = $response;
188 >        print $sock "FQDN\n";
189 >        $response = <$sock>;
190 >        if (!$response || $response eq "ERROR\n") {
191 >            print "The i-scream server did not tell us our FQDN.\n";
192 >            close($sock);
193 >            wait_then_retry();
194 >            next;
195 >        }
196 >        chomp $response;
197 >        $fqdn = $response;
198  
199 <    print $sock "TCPUpdateTime\n";
200 <    $response = <$sock>;
201 <    chop $response;
202 <    $tcp_update_time = $response;
199 >        print "FQDN returned: $fqdn\n";
200 >
201 >        print $sock "UDPUpdateTime\n";
202 >        $response = <$sock>;
203 >        if (!$response || $response eq "ERROR\n") {
204 >            print "The i-scream server did not give us a UDPUpdateTime.\n";
205 >            close($sock);
206 >            wait_then_retry();
207 >            next;
208 >        }
209 >        chomp $response;
210 >        $udp_update_time = $response;
211 >
212 >        print $sock "TCPUpdateTime\n";
213 >        $response = <$sock>;
214 >        if (!$response || $response eq "ERROR\n") {
215 >            print "The i-scream server did not give us a TCPUpdateTime.\n";
216 >            close($sock);
217 >            wait_then_retry();
218 >            next;
219 >        }
220 >        chomp $response;
221 >        $tcp_update_time = $response;
222      
223 <    print "UDP packet period: $udp_update_time seconds.\nTCP heartbeat period: $tcp_update_time.\n";
223 >        print "UDP packet period: $udp_update_time seconds.\nTCP heartbeat period: $tcp_update_time seconds.\n";
224  
225 <    print $sock "ENDCONFIG\n";
226 <    $response = <$sock>;
227 <    chomp $response;
228 <    if (!$response eq "OK") {
229 <        print "ENDCONFIG command to server failed.  Terminated.\n";
230 <        exit(1);
231 <    }
225 >        print $sock "ENDCONFIG\n";
226 >        $response = <$sock>;
227 >        if ($response && !($response eq "OK\n")) {
228 >            print "ENDCONFIG command to server failed.  Terminated.\n";
229 >            close($sock);
230 >            wait_then_retry();
231 >            next;
232 >        }
233  
234 <    print "Config ended.\n";
234 >        print "Config ended.\n";
235      
236 <    print $sock "FILTER\n";
237 <    $response = <$sock>;
238 <    chop $response;
239 <    $response =~ /(.*);(.*);(.*)/;
240 <    ($filter_addr, $udp_port, $tcp_port) = ($1, $2, $3);
236 >        print $sock "FILTER\n";
237 >        $response = <$sock>;
238 >        if (!$response) {
239 >            print "Failed: Could not get a filter address from the filter manager.\n";
240 >            close($sock);
241 >            wait_then_retry();
242 >            next;
243 >        }
244 >        chomp $response;
245 >        if ($response eq "ERROR") {
246 >            print "There are no active configured filters for your host.\n";
247 >            close($sock);
248 >            wait_then_retry();
249 >            next;
250 >        }
251 >        $response =~ /^(.*);(.*);(.*)/;
252 >        ($filter_addr, $udp_port, $tcp_port) = ($1, $2, $3);
253 >        unless (defined($filter_addr) && defined($udp_port) && defined($tcp_port)) {
254 >            print "Failed: Filter address response from server did not make sense: $response\n";
255 >            close($sock);
256 >            wait_then_retry();
257 >            next;
258 >        }
259      
260 <    print "Got filter data ($filter_addr, $udp_port, $tcp_port)\n";
260 >        print "Got filter data ($filter_addr, $udp_port, $tcp_port)\n";
261  
262 <    print $sock "END\n";
263 <    $response = <$sock>;
264 <    chop $response;
265 <    if ($response eq "OK") {
266 <        print "Host successfully configured via TCP.\n"
267 <    }
268 <    else {
269 <        print "The server failed the host configuration on the END command.";
270 <        exit(1);
271 <    }
262 >        print $sock "END\n";
263 >        $response = <$sock>;
264 >        if ($response && ($response eq "OK\n")) {
265 >            print "Host successfully configured via TCP.\n"
266 >        }
267 >        else {
268 >            print "The server failed the host configuration on the END command.\n";
269 >            close($sock);
270 >            wait_then_retry();
271 >            next;
272 >        }
273      
274 <    close($sock);
274 >        close($sock);
275  
276 <    print "Configuration finished sucessfully!\n";
276 >        print "Configuration finished sucessfully!\n";
277 >        last;
278 >    }
279 >    return;
280   }
281  
282 +
283 + #-----------------------------------------------------------------------
284 + # send_udp_packet
285 + # Sends a UDP packet to an i-scream filter.
286 + # The packet contains XML markup describing some of the machine's state.
287 + # Receipt of UDP packets is not guaranteed.
288 + #-----------------------------------------------------------------------
289   sub send_udp_packet() {
290  
291 <    my(@statgrab) = `./statgrab.pl`;
292 <    my(%packet);
293 <    for (my($i) = 0; $i < $#statgrab; $i++) {
294 <        $statgrab[$i] =~ /^([^\s]*) (.*)$/;
295 <        $packet{$1} = $2;
291 >    my($plugins_dir) = "plugins";
292 >    
293 >    opendir PLUGINS, $plugins_dir;
294 >    my(@plugins) = readdir PLUGINS;
295 >    foreach my $plugin (@plugins) {
296 >        push @data, `$plugins_dir/$plugin $ostype` if -x "$plugins_dir/$plugin" && -f "$plugins_dir/$plugin";
297      }
298      
299 +    # get some extra data
300      my($date) = time;
301 +    my($ip);
302 +    $ip = inet_ntoa(scalar(gethostbyname(hostname())) || 'localhost') or $ip = 'localhost';
303      
304 <    my($disk_info) = "<disk>";
305 <    my($i) = 0;
306 <    while (defined $packet{"packet.disk.p$i.attributes.mount"}) {
307 <        $disk_info .= "<p$i>";
308 <        $disk_info .= qq/<name>$packet{"packet.disk.p$i.attributes.name"}<\/name>/;
309 <        $disk_info .= qq/<kbytes>$packet{"packet.disk.p$i.attributes.kbytes"}<\/kbytes>/;
310 <        $disk_info .= qq/<used>$packet{"packet.disk.p$i.attributes.used"}<\/used>/;
164 <        $disk_info .= qq/<avail>$packet{"packet.disk.p$i.attributes.avail"}<\/avail>/;
165 <        $disk_info .= qq/<mount>$packet{"packet.disk.p$i.attributes.mount"}<\/mount>/;
166 <        $disk_info .= "</p$i>";
167 <        ++$i;
168 <    }
169 <    $disk_info .= "</disk>";
170 <
171 <    my($hostname) = hostname();
172 <    $hostname =~ s/\..*$//g;
173 <    my($domainname) = `cat /etc/resolv.conf`;
174 <    $domainname =~ s/domain\s*([^\s]*?)/$1/;
175 <    my($machine_name) = "$hostname.$domainname";
176 <    my($ip) = inet_ntoa(scalar(gethostbyname($hostname)) || 'localhost');
177 <
178 <    # Build the XML packet this way, as we can clearly
179 <    # see the structure and contents... I like this ;-)
180 <    my($xml) = <<EOF;
304 >    # add some extra data to the array
305 >    push(@data, "packet.attributes.seq_no $seq_no");
306 >    push(@data, "packet.attributes.machine_name $fqdn");
307 >    push(@data, "packet.attributes.date $date");
308 >    push(@data, "packet.attributes.type data");
309 >    push(@data, "packet.attributes.ip $ip");
310 >    push(@data, "packet.attributes.key $key");
311      
312 < <packet seq_no="$seq_no" machine_name="$machine_name" date="$date" type="data" ip="$ip">
313 <    <load>
184 <        <load1>$packet{"packet.load.load1"}</load1>
185 <        <load5>$packet{"packet.load.load5"}</load5>
186 <        <load15>$packet{"packet.load.load15"}</load15>
187 <    </load>
188 <    <os>
189 <        <name>$packet{"packet.os.name"}</name>
190 <        <release>$packet{"packet.os.release"}</release>
191 <        <platform>$packet{"packet.os.platform"}</platform>
192 <        <sysname>$packet{"packet.os.sysname"}</sysname>
193 <        <version>$packet{"packet.os.version"}</version>
194 <    </os>
195 <    <users>
196 <        <count>$packet{"packet.users.count"}</count>
197 <        <list>$packet{"packet.users.list"}</list>
198 <    </users>
199 <    <processes>
200 <        <total>$packet{"packet.processes.total"}</total>
201 <        <sleeping>$packet{"packet.processes.sleeping"}</sleeping>
202 <        <zombie>$packet{"packet.processes.zombie"}</zombie>
203 <        <stopped>$packet{"packet.processes.stopped"}</stopped>
204 <        <cpu>$packet{"packet.processes.cpu"}</cpu>
205 <    </processes>
206 <    <cpu>
207 <        <idle>$packet{"packet.cpu.idle"}</idle>
208 <        <user>$packet{"packet.cpu.user"}</user>
209 <        <kernel>$packet{"packet.cpu.kernel"}</kernel>
210 <        <iowait>$packet{"packet.cpu.iowait"}</iowait>
211 <        <swap>$packet{"packet.cpu.swap"}</swap>
212 <    </cpu>
213 <    <memory>
214 <        <total>$packet{"packet.memory.real"}</total>
215 <        <free>$packet{"packet.memory.free"}</free>
216 <    </memory>
217 <    <swap>
218 <        <total>$packet{"packet.memory.swap_total"}</total>
219 <        <free>$packet{"packet.memory.swap_free"}</free>
220 <    </swap>
221 <    $disk_info
222 < </packet>
312 >    # sort the data
313 >    @data = sort(grep(!/^$/, grep(/^packet\./, @data)));
314  
315 < EOF
315 >    # turn the array into some nice XML
316 >    my($xml) = &make_xml("", "");
317  
226    $xml =~ s/\n\s*//g;
227    
318      my($sock) = new IO::Socket::INET (
319                                        PeerPort => $udp_port,
320                                        PeerAddr => $filter_addr,
321                                        Proto => 'udp'
322 <                                     ) or die "Socket: $!\n";
323 <
322 >                                     ) or die "Could not send UDP: $!\n";
323 >    
324      print $sock $xml or die "Could not send UDP packet: $!\n";
325      close($sock);
326      $seq_no++;
327      print "-";
328 +    
329 +    return;
330   }
331  
332 +
333 + #-----------------------------------------------------------------------
334 + # send_tcp_heartbeat
335 + # Establishes a TCP connection to an i-scream filter.
336 + # The heartbeat is used as a guaranteed "I'm alive" delivery mechanism.
337 + # If we need to reconfigure, then we complete the heartbeat before
338 + # doing so.
339 + #-----------------------------------------------------------------------
340   sub send_tcp_heartbeat() {
341  
342 +    my ($doReconfigure) = 0;
343 +
344      my($sock) = new IO::Socket::INET(
345                                       PeerAddr => $filter_addr,
346                                       PeerPort => $tcp_port,
347                                       Proto => 'tcp'
348 <                                    ) or die "Could not perform heartbeat via TCP: $!\n";
348 >                                    ) or return;
349 >    if (!defined $sock) {
350 >        print "IHOST WARNING: Failed to deliver a heartbeat to the i-scream filter.\n";
351 >        &tcp_configure();
352 >        return;
353 >    }
354  
248    die "Could not connect to the i-scream filter: $!\n" unless $sock;
249
355      # Now run through the configuration process.
356      my($response);
357  
358      print $sock "HEARTBEAT\n";
359      $response = <$sock>;
360 <    chop $response;
256 <    if (!$response eq "OK") {
360 >    if (!$response eq "OK\n") {
361          close($sock);
362          print "Server gave wrong response to HEARTBEAT: $response\n";
363 +        &tcp_configure();
364          return;
365      }
366      
367      print $sock "CONFIG\n";
368      $response = <$sock>;
369 <    chop $response;
265 <    if (!$response eq "OK") {
369 >    if (!$response eq "OK\n") {
370          close($sock);
371          print "Server gave wrong response to CONFIG: $response\n";
372 +        &tcp_configure();
373          return;
374      }
375  
376      print $sock "$file_list\n";
377      $response = <$sock>;
378 <    chop $response;
274 <    if (!$response eq "OK") {
378 >    if (!$response eq "OK\n") {
379          close($sock);
380          print "Server gave wrong response to file list: $response\n";
381 +        &tcp_configure();
382          return;
383      }
384      
385      print $sock "$last_modified\n";
386      $response = <$sock>;
387 <    chop $response;
283 <    if ($response eq "ERROR") {
387 >    if ($response eq "ERROR\n") {
388          close($sock);
389 <        &tcp_configure();
390 <        return;
389 >        print "Server configuration changed.  Reconfiguring with filter manager.\n";
390 >        $doReconfigure = 1;
391      }
392 <    if (!$response eq "OK") {
392 >    if (!$response eq "OK\n") {
393          close($sock);
394          print "Server gave wrong response to HEARTBEAT: $response\n";
395 +        &tcp_configure();
396          return;
397      }
398      
399 +    print $sock "KEY\n";
400 +    $key = <$sock>;
401 +    
402      print $sock "ENDHEARTBEAT\n";
403      $response = <$sock>;
404 <    chop $response;
297 <    if (!$response eq "OK") {
404 >    if (!$response eq "OK\n") {
405          close($sock);
406          print "Server gave wrong response to ENDHEARTBEAT: $response\n";
407 +        &tcp_configure();
408          return;
409      }
410      
411      close($sock);
412      print "^";
413 +    
414 +    &tcp_configure() if $doReconfigure;
415 +    
416 +    return;
417 + }
418 +
419 +
420 + #-----------------------------------------------------------------------
421 + # write_pid
422 + # Writes the PID (process ID) of this instance to $pidfile.
423 + # This is then used by a seperate script to check (and restart) ihost.
424 + #-----------------------------------------------------------------------
425 + sub write_pid() {
426 +    open PID, ">$pidfile";
427 +    print PID $$;
428 +    close PID;
429 +    
430 +    return;
431 + }
432 +
433 + #-----------------------------------------------------------------------
434 + # make_xml
435 + # Turns an array of plugins data into an XML string.
436 + #-----------------------------------------------------------------------
437 + sub make_xml() {
438 +    my($curlevel, $curline) = @_;
439 +    my($xmltemp) = ""; my($curtag) = ""; my($attributes) = "";
440 +    while(1) {
441 +        $curline = shift(@data) if $curline eq "";
442 +        return $xmltemp if not defined $curline;
443 +        chomp $curline;
444 +        # dealing with nest (or attributes)
445 +        if($curline =~ /^$curlevel([^\.\s]+\.)/) {
446 +            $curtag=$1;
447 +            if($curline =~ /^$curlevel$curtag([^\.\s]+)\s+(.*)$/) {
448 +                $xmltemp .= &make_xml("$curlevel$curtag", $curline);
449 +            }
450 +            elsif($curline =~ /^$curlevel$curtag(attributes)\.([^\.\s]+)\s+(.*)$/) {
451 +                $attributes .= " $2=\"$3\"";
452 +            }
453 +            else {
454 +                $xmltemp .= &make_xml("$curlevel$curtag", $curline);
455 +            }
456 +            my($nextline) = $data[0]; chomp $nextline if defined $nextline;
457 +            $curtag =~ s/(.*)\./$1/;
458 +            if((defined $nextline) && ($nextline =~ /^$curlevel$curtag\./)) {
459 +                $curline = "";
460 +            }
461 +            else {
462 +                $xmltemp = "<$curtag$attributes>$xmltemp</$curtag>" unless $curtag eq "";
463 +                return $xmltemp;
464 +            }
465 +        }
466 +        # dealing with value
467 +        elsif($curline =~ /^$curlevel([^\.\s]+)\s+(.*)$/) {
468 +            $curtag=$1;
469 +            $xmltemp=$2;          
470 +            my($nextline) = $data[0]; chomp $nextline if defined $nextline;
471 +            if(defined $nextline && ($nextline =~ /^$curlevel$curtag\./ || $nextline =~ /^$curlevel$curtag\s+/)) {
472 +                $curline = "";
473 +            }
474 +            else {
475 +                $xmltemp = "<$curtag$attributes>$xmltemp</$curtag>" unless $curtag eq "";
476 +                return $xmltemp;
477 +            }
478 +        }
479 +        # dealing with a null value
480 +        elsif($curline =~ /^$curlevel([^\.\s]+)$/) {
481 +            # simply adding a space makes the above elsif deal with it :)
482 +            # just level with an empty tag in the XML
483 +            $curline .= " ";
484 +        }
485 +        # failing all that, skip the line
486 +        else {
487 +            $curline = "";
488 +        }
489 +    }
490   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines